Coupling different machine learning and meta-heuristic optimization techniques to generate the snow avalanche susceptibility map in French Alps

Abstract

Snow avalanches are one of the most damaging natural disasters which lead to serious disruptions in landscape, transportation, and most importantly human life. Therefore, predetermination of the regions susceptible to such incidents is a crucial task to mitigate their adverse impacts. Most of the studies conducted to generate snow avalanche susceptibility mapping employ various machine learning (ML) algorithms as they offer both computational advances and high predictive success. Thus, the focus of this thesis is to introduce a hybrid predictive framework encompassing the application of different meta-heuristic optimization techniques and ML algorithms. To accomplish this aim, the present thesis sought the acquire the best performed model among 9 different hybrid scenarios encompassing three different meta-heuristics, namely particle swarm optimization (PSO), gravitational search algorithm (GSA), and cuckoo search (CS), and three different ML approaches, i.e., support vector classification (SVC), stochastic gradient boosting (SGB), and k-nearest neighbors (KNN), pertaining to different predictive families. According to diligent analysis performed with regard to the blinded testing set, the PSO-SGB illustrated the most satisfactory predictive performance with an accuracy of 0.815, while the precision and recall were found as 0.824 and 0.821, respectively. The F1-score of the predictions was found as 0.821 and the area under receiver operating curve (AUROC) was obtained as 0.9. Despite attaining similar predictive success via the CS-SGB model, the time-efficiency analysis underscored the capability of the PSO-SGB as the corresponding process consumed considerably lower computational time compared to its counterpart. In line with the respective outcomes, the snow avalanche susceptibility map pertaining to the focalized area was generated. It is especially worth noting that the current thesis further incorporated the SHapley Additive exPlanations (SHAP) into the predictive framework to post-process the attained findings, making the established hybrid model more explainable. Finally, slope, elevation, and wind speed were found as the most contributing attributes to the detection of regions having susceptibility to the snow avalanche incidents. The results are expected to assist in decision-makers in French Alps to not only address the regions that potentially experience snow avalanches but also identify the mitigation measures aiding to minimize adverse consequences of corresponding disasters.